A grid-less flexible tactile sensing system based on deep neural network for two-point localization and shape recognition

This research proposes a grid-less planar flexible sensor that can simultaneously sense the number, shape, magnitude and localization of pressure at touch points with high precision. With the help of deep neural network (DNN), ultra-precise sensing can be realized with much fewer electrodes and much concise signal processing than traditional sensor arrays, offering a practical solution for low-cost and high-precision flexible tactile sensing systems. Test results indicate that the DNN-based grid-less planar flexible tactile sensor sample had ≤4% prediction error for both pressure magnitude and localization for two-point sensing. Additionally, the localization accuracy was >10 times higher than traditional sensor arrays with the same number of electrodes. At the same time, the sensor can also distinguish 16 different shapes of touch with an accuracy of 97.7%, even though these shapes involve different areas, and some of which are difficult to distinguish by the human tactile sense.